@axtary/config
v0.6.1
Published
YAML configuration loading for local Axtary runtime packages.
Maintainers
Readme
@axtary/config
YAML configuration loading for local Axtary runtime packages.
Early 0.x release: the runtime path is real and tested, but the API is not stable yet and may change between minor versions.
The source repository is currently private. Public product documentation and runnable guides are at axtary.com/docs.
npm install @axtary/configWhat It Does
- Parses
axtary.ymltext into a typed runtime config. - Applies defaults for issuer, tenant, ledger path, runtime mode, adapters, and policy.
- Normalizes policy through
@axtary/policy. - Resolves ledger paths relative to the config file or current working directory.
- Emits a default config template for demos.
- Provides
CachedConfigLoaderfor long-running local proxy sessions.
Adapter Config
GitHub defaults to credential-free demo mode (the v0 config literal is fake):
adapters:
github:
mode: fakeTo route allowed GitHub actions to the real REST API, set:
adapters:
github:
mode: rest
tokenEnv: GITHUB_TOKEN
apiBaseUrl: https://api.github.com
userAgent: axtary-local-proxySlack and Linear follow the same credential-free demo-first pattern:
adapters:
slack:
mode: web
tokenEnv: SLACK_BOT_TOKEN
apiBaseUrl: https://slack.com/api
linear:
mode: graphql
tokenEnv: LINEAR_API_KEY
apiUrl: https://api.linear.app/graphqlLinear and Jira schemas/defaults are consumed from the same native-connector governance descriptors that drive evidence and doctor metadata, so a connector cannot silently drift between config and operator guidance.
Jira Cloud supports API-token and OAuth-backed REST modes:
adapters:
jira:
mode: rest
auth: api_token
emailEnv: JIRA_EMAIL
tokenEnv: JIRA_API_TOKEN
siteUrl: https://your-site.atlassian.netFor OAuth, set auth: oauth; axtary connect jira stores the access/refresh
tokens and discovered cloud ID in the local credential broker.
Google Drive uses one-file Picker OAuth and a separate policy allowlist:
adapters:
drive:
mode: rest
maxReadBytes: 100000
policy:
drive:
reads:
allowedFileIds: [YOUR_PICKED_FILE_ID]
maxReadBytes: 100000Run axtary connect drive --redirect-uri https://<tunnel>/callback; the
drive.file token and selected file id stay in the local credential broker.
The CLI reads tokens from the named environment variables. Keep credential-free demo mode for demos and tests; switch to the implemented live mode when you want provider calls.
Local docs are enabled by default and resolve roots relative to the config file:
adapters:
docs:
mode: local
workspace: local
roots:
- docs
- product
maxSearchResults: 10
maxReadBytes: 20000Pair this with policy.docs.documents to scope allowed workspaces, readable path prefixes, denied prefixes, and per-action limits.
Multi-rule policy block
The legacy policy: section remains the built-in preset's configuration.
Define policies: to activate the named rule engine:
policies:
defaultDecision: deny
rules:
- id: allow-local-docs
match:
tools: docs.documents.*
resources: docs:*
effect: allow
obligations:
expiresInSeconds: 120
blockedPathPrefixes: [private/, secrets/]See examples/axtary.policies.yml for a complete multi-rule configuration.
Runtime Latency Guard
runtime:
handlerTimeoutMs: 10000Handlers that exceed the timeout fail closed with proxy_handler_timeout:<ms>ms. Proxy responses include per-stage timings so local runs can track policy, signing, ledger, and adapter latency.
Runtime budgets
Budget enforcement is disabled by default. Limits and costs are authority-owned configuration:
runtime:
budget:
enabled: true
scope: tenant # tenant | actor | task
limits:
actions: 100
externalMessages: 10
costs:
"*":
actions: 1
slack.chat.postMessage:
externalMessages: 1Wildcard and exact-tool costs are additive. An enabled budget requires the
"*" cost and every cost dimension must have a limit; incomplete trusted
configuration fails at runtime startup. This is local metering for enforcement
and ledger evidence, not pricing or billing configuration.
ActionPass status distribution
ActionPass v2 status distribution is enabled by default:
runtime:
status:
enabled: true
keyringPath: .axtary/keys/actionpass-issuer-keyring.json
storePath: .axtary/actionpass-status.json
ttlSeconds: 60
expiresInSeconds: 120The keyring and status store are private local authority state. The proxy publishes public JWKS and a signed freshness-bounded status list; verifiers fail closed when required status is unavailable, stale, or invalid.
Generic SSF/CAEP ingestion is opt-in:
runtime:
caep:
enabled: true
issuer: https://transmitter.example
audience: https://local.axtary.dev/ssf/events
jwksUri: https://transmitter.example/.well-known/jwks.json
mappingPath: .axtary/caep-mappings.json
replayPath: .axtary/ssf-replay.jsonWhen enabled, issuer, audience, and JWKS URI are required at startup. An
optional bearerTokenEnv adds a shared transport credential without placing
its value in config. Signed CAEP subjects still require explicit caep-map
root-pass mappings; Axtary does not infer identity-provider associations.
Cached Loading And Hot Reload
CachedConfigLoader caches parsed configs and checks the config file mtime on getConfig(). File-backed configs reload automatically when axtary.yml changes, which lets the local proxy pick up policy edits without a restart.
Deterministic provenance enforcement is opt-in under the policy block:
policy:
provenance:
enabled: true
unknownDecision: deny
untrustedDecision: step_upEnabling it without a trusted runtime provenance resolver intentionally causes
covered egress fields to fail closed as provenance_unknown_egress.
If a file-backed config disappears or becomes unreadable, the loader throws instead of silently serving stale policy. The proxy records that as a fail-closed policy-resolution denial.
Quickstart
This example runs as-is with Node 20+:
import { parseAxtaryConfigYaml, DEFAULT_AXTARY_CONFIG_YAML } from "@axtary/config";
// Parse and validate an axtary.yml (here: the scaffold `axtary init` writes).
const config = parseAxtaryConfigYaml(DEFAULT_AXTARY_CONFIG_YAML);
console.log(config.policy ? "policy loaded" : "no policy", Object.keys(config));Design Notes
Config parsing is intentionally separate from the CLI so adapters, tests, and future MCP wrappers can all share the same file format.
